Researchers at Georgia Institute of Technology have developed an artificial intelligence that creates video games simply by watching gameplay footages.
The AI system developed by Georgia Institute of Technology researchers was made to watch video games such as Super Mario Bros., Mega Man and Sonic the Hedgehog. After watching a video game in less than two minutes, the AI then creates its own video game – creating on its own the basic software of the game that governs everything from character movements to rendering graphics.
“Our AI creates the predictive model without ever accessing the game’s code, and makes significantly more accurate future event predictions than those of convolutional neural networks,” Matthew Guzdial, lead researcher of the AI system, said in a statement.
Mark Riedl, associate professor of Interactive Computing and co-investigator on the AI system, said that this is the first AI technique that learns a video game and develops a video game based on gameplay footage. “The technique relies on a relatively simple search algorithm that searches through possible sets of rules that can best predict a set of frame transitions,” Riedl said.
The AI system developed by the Georgia Institute of Technology researchers, however, cannot create a video game based on games with off-screen actions such as Clash of Clans as the system works well only with games where the action happens on-screen.
According to the researchers, this AI system can be used by game studios to speed up game development. It can also be used by game enthusiasts to experiment with different styles of play based on existing games.